Can correlation be used as a measure of accuracy?

I have a paired data set of known age and estimated age, and correlation tests show that there is a strong correlation between the two, but what I want to know is, does that mean the estimates are accurate?

My thinking is no, correlation cannot tell you if the estimated age is accurate, but I'm not 100%.

Additional info: the estimated ages are obtained by following a method that examines features of the skeleton.

Thanks in advance!
I agree that correlation can't tell you that. After all, there would be a perfect correlation of r=1.0 between the following pairs of estimated age/actual age: 3/83, 4/84, 5/85, 6/86, 7/87. (But as you can see the guesses are all 80 years off the mark).

I feel like either a standard deviation of the difference scores or a mean absolute difference score would be a pretty good metric of accuracy in this case. But I have no idea if either of those are actually done in cases like this.

So in terms of approaches that do work, I'll let others who know what they're talking about more than me point you in the best direction.
Last edited:


Less is more. Stay pure. Stay poor.
Yeah, I never really used it, but this seems like what Bland-Altman could be used for. Not familiar with Youden plot. I know Youden Index, but that is for a binary by continuous comparison (I least I thought). Probably won't hurt put them in a scatterplot or a histogram of the differences.


TS Contributor


Less is more. Stay pure. Stay poor.

Tell me more about the plot, please. So you have to have two paired continuous values. How is this different from a correlation. I am too tired to figure this out by reading on the web. I need some one to spoon feed me knowledge.


TS Contributor
It is essentially a simple graphical approach to determine whether the between group variation is greater than the within group variation with paired data. It's initial application was comparing inter-laboratory test results on the same items. It helps to determine whether labs are equivalent/biased, detect differences in variability and outliers.